We examined the relationship between the recovery of hand and arm function in a group of hemiplegic stroke patients and the presence of short-latency EMG responses to transcranial magnetic stimulation (TMS) in 4 different upper limb muscles (deltoid, biceps, extensor digitorum communis and the first dorsal interosseous). Twenty-one patients were examined within 5 weeks of stroke (median 2 weeks), and then at regular intervals over the next 12 months. Some patients recovered rapidly (Group A); in others, recovery was slow and incomplete (Group B). Even at the first test, Group A patients had responses to TMS in all muscles. Most Group B patients initially lacked responses in all tested upper limb muscles; in those that later were able to activate hand muscles, responses returned at or just before this stage of recovery. No such clear correlation between the presence of responses to TMS and ability to activate more proximal arm muscles was evident. Response latency was initially long and declined in a manner that was highly correlated with muscle strength and hand function test scores. Ipsilateral responses were elicited from both the affected and unaffected hemispheres. Ipsilateral responses from the latter were most common in the proximal muscles of the affected limb, and had latencies that were longer than those elicited in the contralateral (unaffected) arm. Nine cases of ipsilateral responses in hand muscles were found; such responses are not found in healthy subjects. Ipsilateral responses from the undamaged hemisphere were more prevalent in the poorly recovered patients; the underlying mechanisms may not be beneficial for recovery.